Open Role
Founding ML Engineer
RemoteUS / JPFull-Time
Neognathae is Auxerta's from-scratch model architecture — a system of domain-specialized models, each trained to be an expert in its vertical. Rather than building one general-purpose model, we are building a coordinated set of models where each one deeply understands a specific domain: law, medicine, engineering, finance, and beyond.
This is a founding role. You will work at the architecture level — designing training pipelines, shaping how models specialize, and building the infrastructure that makes a fleet of expert models operate as a unified system. You'll work directly with the founders and have outsized ownership over the technical direction of Neognathae.
What You'll Work On
- Training from-scratch models across specialized domains with custom architectures and training loops
- Building and optimizing data pipelines for curating high-fidelity, domain-specific training corpora
- Designing evaluation frameworks and benchmarks that measure genuine domain expertise, not surface fluency
- Developing routing and orchestration layers that coordinate multiple expert models into a coherent system
- Inference optimization — quantization, serving, and latency reduction for production deployment
- Agent and tool integrations that enable expert models to interface with domain-specific APIs, databases, and workflows
What We're Looking For
- Experience training language models from scratch — not just fine-tuning, but architecture decisions, loss functions, and training dynamics
- Deep proficiency in PyTorch or JAX; hands-on with distributed training (FSDP, DeepSpeed, or equivalent)
- Strong understanding of data quality — you've built pipelines that filter, deduplicate, and validate training data at scale
- Experience designing and running rigorous evaluations: perplexity is not enough, you care about domain-specific accuracy
- Familiarity with mixture-of-experts architectures, model merging, or multi-model orchestration is a strong plus
- Self-directed, low-ego, high-output — you thrive with ambiguity and move fast at an early-stage company
Important: Candidates must be authorized to work in the United States or Japan without employer sponsorship. Auxerta does not sponsor employment visas at this time. This role requires the ability to travel to Japan periodically.